Self-adaptive control method for range hood and range hood

Abstract

A range hood adaptive control method and a range hood, the method comprising: acquiring environmental data, the environmental data comprising gas data for characterizing the concentration of at least one to-be-detected gas component generated by cooking, and the environmental temperature and environmental humidity of a cooking area; the environment data is input into a preset machine learning model, a prediction result is obtained, and the preset machine learning model is used for predicting the current cooking stage and the smoke amount generated by cooking according to the environment data; and adjusting the working state of the range hood according to the prediction result. According to the scheme, the working state of the range hood can be automatically, intelligently and accurately adjusted according to the current cooking environment, the overall performance of the range hood can actively meet the real-time cooking requirement of a user without manual adjustment of the user, and the safety of the cooking process is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of household appliances, and particularly to an adaptive control method for a range hood and a range hood. Background Art

[0002] During daily cooking, users need to manually adjust the exhaust intensity of the range hood according to the amount of oil fume generated during current cooking. If the adjustment is too frequent, it is very easy to get flustered and fail to attend to all matters. For example, if the exhaust intensity is not increased in time when the oil fume is large, excessive oil fume will cause the user to cough; if the cooking heat is not controlled in time in order to adjust the range hood, the food will be overcooked and can only be discarded, and in severe cases, the cooking utensils will be burned out.

[0003] In view of the above problems, some range hoods are equipped with an automatic exhaust intensity adjustment function, which determines the amount of oil fume based on gas data and then adjusts the exhaust intensity of the range hood accordingly. However, the current automatic adjustment mechanism of range hoods usually judges whether there is a sharp change in the amount of oil fume according to the change rate of the data collected by the gas sensor, and then adjusts the fan speed of the range hood accordingly. The judgment accuracy of this automatic adjustment mechanism is very low, resulting in untimely and inaccurate adjustment of the exhaust intensity. On the other hand, existing range hoods cannot detect whether overcooking has occurred, and thus cannot make the most appropriate response to this phenomenon, and cannot effectively avoid food waste and damage to cooking utensils, affecting the safety of the cooking process. Summary of the Invention

[0004] An object of the embodiments of the present invention is to provide an improved adaptive control method for a range hood and a range hood.

[0005] Therefore, the embodiments of the present invention provide an adaptive control method for a range hood, including: obtaining environmental data, where the environmental data includes gas data for characterizing the concentration of at least one gas component generated by cooking, the ambient temperature and ambient humidity of the cooking area; inputting the environmental data into a preset machine learning model and obtaining a prediction result, where the preset machine learning model is used to predict the current cooking stage and the amount of oil fume generated by cooking according to the environmental data; and adjusting the working state of the range hood according to the prediction result.

[0006] In the prior art, the working state of the range hood is mainly adjusted manually by the user, or simply adjusted when the gas data changes suddenly, resulting in untimely and inaccurate adjustment of the working state of the range hood. In contrast, the present embodiment automatically and intelligently adjusts the working state of the range hood precisely according to the current cooking environment, and can actively make the overall performance of the range hood meet the user's real-time cooking needs without manual adjustment by the user, improving the safety of the cooking process. Specifically, the smoke volume and the current cooking stage are determined comprehensively based on the gas data and the ambient temperature and humidity data to more precisely adjust the working state of the range hood. Among them, the working state may include the operating parameters of the fan, and the current cooking stage may include the normal cooking stage and the overcooking stage. Thus, the fan speed can be controlled more precisely, which is beneficial to preventing food waste and burning of cooking utensils, making the cooking process safer.

[0007] Optionally, the preset machine learning model includes: a first model for predicting the smoke volume according to the environmental data; a second model for predicting the current cooking stage according to the environmental data; and the step of inputting the environmental data into the preset machine learning model and obtaining the prediction result includes: inputting the environmental data into the first model to obtain a first prediction result, where the first prediction result includes the prediction result of the smoke volume; inputting the environmental data into the second model to obtain a second prediction result, where the second prediction result includes the prediction result of the current cooking stage. Thus, the preset machine learning model combines the smoke volume prediction algorithm and the overcooking detection algorithm, and can precisely control the working state of the range hood and realize timely monitoring of overcooking according to the prediction result. Specifically, the first model is dedicated to predicting the size of the smoke volume to achieve real-time detection of the smoke volume, so as to precisely control the working state of the range hood. Further, the second model is dedicated to predicting the current cooking stage to achieve timely detection of the overcooking stage.

[0008] Optionally, the step of adjusting the working state of the range hood according to the prediction result includes: if the first prediction result indicates that the smoke volume generated by cooking is zero, adjusting the working state of the range hood to the shutdown or standby state; if the first prediction result indicates that the smoke volume generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the normal cooking stage, adjusting the fan speed in the range hood according to the first prediction result; if the first prediction result indicates that the smoke volume generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the overcooking stage, controlling the fan of the range hood to operate at the maximum speed. Thus, combining the first prediction result and the second prediction result comprehensively predicts the real-time cooking environment to determine the most suitable working state of the range hood at present, and then precisely controls the fan speed so that the range hood works in the most suitable working state for the current cooking environment.

[0009] Optionally, the first model and the second model are constructed using different machine learning algorithms. Thus, the first model and the second model can be independently constructed and trained so that the prediction results of each model conform to their respective model construction objectives. For example, models are constructed using multiple machine learning algorithms respectively, and the model with the highest prediction accuracy for the smoke volume is determined as the first model. Similarly, models are constructed using multiple machine learning algorithms respectively, and the model with the highest prediction accuracy for the current cooking stage is determined as the second model. When selecting appropriate models as the first model and the second model respectively, there is no correlation in the selection logic. Thus, constructing the first model and the second model respectively based on the prediction result accuracy is beneficial to improving the accuracy of the first prediction result and the second prediction result.

[0010] Optionally, the training process of the preset machine learning model includes: obtaining a training set, where the training set includes environmental data and corresponding calibration results obtained when different cooking methods are used for different ingredients and at multiple cooking stages. The multiple cooking stages at least include a normal cooking stage and an overcooking stage, and the calibration results include a standard smoke volume and a standard cooking stage; training the preset machine learning model based on the training set until the prediction accuracy of the preset machine learning model reaches a preset threshold. Thus, by enriching the diversity of training samples in the training set, the preset machine learning model can also have better performance in complex environments and can accurately identify the smoke volume and the current cooking stage for different ingredients and different cooking stages.

[0011] Optionally, the preset machine learning model includes: a first model for predicting the smoke volume according to the environmental data; a second model for predicting the current cooking stage according to the environmental data; the training process of the first model includes: obtaining a first set, where the first set includes environmental data and corresponding standard smoke volumes obtained when different cooking methods are used for different ingredients and at multiple cooking stages; training the first model based on the first set until the prediction accuracy of the first model reaches a first threshold; the training process of the second model includes: obtaining a second set, where the second set includes environmental data and corresponding standard cooking stages obtained when different cooking methods are used for different ingredients and at multiple cooking stages; training the second model based on the second set until the prediction accuracy of the second model reaches a second threshold. Thus, different sets are constructed corresponding to the specific dimensions of the prediction results of the two models for targeted training, so that the prediction accuracy of the first prediction result of the first model in the smoke volume dimension meets the expectation, and the prediction accuracy of the second prediction result of the second model in the cooking stage dimension meets the expectation. Further, by enriching the diversity of the training samples in the two sets, the first model and the second model can also have better performance in a complex environment, and can accurately identify the smoke volume and the current cooking stage for different ingredients and different cooking stages.

[0012] Optionally, the numerical values of the first threshold and the second threshold are different. Thus, the expectations for the respective prediction accuracies of the two models can be flexibly set according to needs, so that the trained first model and second model meet the user's expectations.

[0013] Optionally, the environmental temperature of the cooking area is determined according to multiple candidate temperatures, and the multiple candidate temperatures are respectively collected from different positions of the cooking area; and / or, the environmental humidity of the cooking area is determined according to multiple candidate humidities, and the multiple candidate humidities are respectively collected from different positions of the cooking area. Thus, the environmental temperature / environmental humidity is comprehensively calculated based on the candidate temperatures / candidate humidities at multiple positions, which is beneficial to improving the detection accuracy of the environmental temperature / environmental humidity. For example, both the range hood and the stove have sensors for temperature / humidity measurement to respectively collect the candidate temperatures / candidate humidities at different positions.

[0014] Therefore, an embodiment of the present invention further provides an oil fume machine, including: a main body; a control module disposed in the main body, where the control module is configured to execute the above-mentioned adaptive control method for the oil fume machine; a collection module disposed in the main body and communicating with the control module, where the collection module is configured to collect the environmental data and transmit it to the control module. Thus, based on the adaptive control method for the oil fume machine provided by this implementation solution, the real-time smoke volume and the current cooking stage can be accurately judged, so that the oil fume machine can intelligently and accurately adjust its own working state, so that the working state of the oil fume machine always conforms to the real-time cooking environment. Further, the oil fume machine can maintain operation in a working state consistent with the real-time cooking environment, which is beneficial to preventing food waste and burning of cooking utensils, making the cooking process safer.

[0015] Optionally, the oil fume machine further includes: a fan disposed in the main body, and the control module adjusts the rotation speed of the fan according to the prediction result. Thus, by combining the automatic fan speed control algorithm and the overcooking detection algorithm, the rotation speed of the fan can be accurately controlled and overcooking detection can be achieved. Specifically, the automatic fan speed control algorithm may include automatically controlling the fan speed based on the predicted smoke volume result of a preset machine learning model, and the overcooking detection algorithm may include detecting whether there is an overcooking phenomenon based on the predicted result of the current cooking stage of a preset machine learning model.

[0016] Optionally, the oil fume machine further includes: a storage module disposed in the main body and communicating with the control module, where the storage module is configured to store the preset machine learning model. Thus, the preset machine learning model is pre-disposed in the oil fume machine and is called when this implementation solution is executed to achieve the prediction of the smoke volume and the current cooking stage.

[0017] Optionally, the control module further communicates with at least one auxiliary sensor, the auxiliary sensor is externally disposed on the oil fume machine and is separately disposed at different positions in the cooking area from the oil fume machine, and the auxiliary sensor is configured to collect at least a part of the environmental data and transmit it to the control module. Thus, the collection result of the auxiliary sensor can reinforce and optimize the collection result of the collection module, so that the environmental data input into the preset machine learning model can more accurately represent the real-time cooking environment. Further, combining the environmental data obtained by the collection module and the auxiliary sensor is beneficial to enriching the dimensional diversity of the input data input into the preset machine learning model. For example, an auxiliary sensor can be disposed on the cooker to collect the environmental temperature / environmental humidity, and these data are input into the preset machine learning model together with the gas data collected by the collection module disposed on the oil fume machine to obtain the prediction result.

[0018] Optionally, the acquisition module includes: a gas detector and a temperature and humidity sensor. Thus, gas data is acquired based on the gas detector, and temperature data and humidity data are acquired based on the temperature and humidity sensor. Description of the Drawings

[0019] Figure 1 is a flowchart of an adaptive control method for an oil fume machine according to an embodiment of the present invention;

[0020] Figure 2 is a schematic diagram of the training process of a preset machine learning model according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of a typical application scenario of an oil fume machine according to an embodiment of the present invention;

[0022] Figure 4 is a logic block diagram of the adaptive control of an oil fume machine in a typical application scenario according to an embodiment of the present invention;

[0023] In the drawings:

[0024] 3 - Oil fume machine; 31 - Body; 32 - Control module; 33 - Acquisition module; 331 - Gas detector; 332 - Temperature and humidity sensor; 34 - Fan; 35 - Communication module; 36 - Storage module; 37 - Display interface; 4 - Stove. Detailed Embodiments

[0025] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be given with reference to the accompanying drawings.

[0026] Figure 1 is a flowchart of an adaptive control method for an oil fume machine according to an embodiment of the present invention.

[0027] This implementation scheme can be applied to the smart home application scenario, specifically to the intelligent control scenario of the oil fume machine. For example, by real - time monitoring the current cooking environment, the working state of the oil fume machine is adaptively adjusted, and the overall performance of the oil fume machine can be intelligently adjusted to a working state that meets the user's real - time cooking needs without manual intervention by the user throughout the process.

[0028] The current cooking environment in this embodiment may include the real - time smoke volume in the cooking area, and the smoke volume may be the oil fume / smoke generated by cooking utensils, stoves, ingredients being cooked, etc. in the cooking area. Further, the current cooking environment may also include the current cooking stage, specifically, it may be the cooking stage of a single dish being cooked in the cooking area, or the overall cooking stage of multiple dishes being cooked on multiple stoves in the cooking area.

[0029] The cooking area in this embodiment may include the area where the cooking appliance is located, and may also include the area that can be covered by the exhaust power of the range hood. In some embodiments, the range hood may be located above the cooking area.

[0030] This implementation scheme can be executed by a control module. The control module can be, for example, the single-chip microcomputer of the range hood, or can also be, for example, the control unit dedicated to executing this implementation scheme in the range hood. In this embodiment, the control module can be arranged in the body of the range hood.

[0031] Specifically, referring to Figure 1 , the method for adaptive control of the range hood in this embodiment may include the following steps:

[0032] Step S101, obtain environmental data, where the environmental data includes gas data for characterizing the concentration of at least one gas component to be detected generated by cooking, the environmental temperature and environmental humidity of the cooking area;

[0033] Step S102, input the environmental data into a preset machine learning model and obtain a prediction result, where the preset machine learning model is used to predict the current cooking stage and the amount of smoke generated by cooking according to the environmental data;

[0034] Step S103, adjust the working state of the range hood according to the prediction result.

[0035] More specifically, the environmental data can be used to characterize the real-time state of the current cooking environment.

[0036] Further, the gas components to be detected may be generated from the ingredients being cooked. Different types of ingredients may generate different types of gas components during cooking, and the same type of ingredient may also generate different types of gas components at different cooking stages. For example, ingredients can include meat, nuts, vegetables, etc. according to types. Different gas components may affect dimensional indexes such as the amount and concentration of the generated smoke. The amount of smoke generated by different cooking actions may also be different.

[0037] Further, the environmental temperature of the cooking area can be used to characterize the temperature change in the cooking area due to the user's cooking actions. For example, the environmental temperature of the cooking area can include the temperature within a preset space area near the cooking appliance, and further can include the temperature of the space area near the cooking utensil. Also for example, the environmental temperature of the cooking area can include the temperature of the ingredients being cooked. The change in the environmental temperature may affect dimensional indexes such as the amount and concentration of the generated smoke. The environmental temperature of the cooking area can be different when the user performs different cooking actions. For example, the environmental temperature during stir-frying is higher than that during boiling water.

[0038] Furthermore, the ambient humidity in the cooking area can be used to characterize the humidity change in the cooking area caused by the user's cooking actions. For example, the ambient humidity can include the humidity near the cooking stove, and further can include the humidity near the cooking appliance. Also for example, the ambient humidity can include the humidity of the ingredients being cooked. The change in ambient humidity may affect dimensional indicators such as the amount and concentration of the generated smoke.

[0039] The ambient temperature and the ambient humidity can be used as compensation values for the gas data and are input into the preset machine learning model as ambient data together. These three types of data can comprehensively reflect the amount of smoke, which is beneficial to improving the accuracy of model prediction.

[0040] Furthermore, the cooking stage can include: a normal cooking stage, an overcooking stage, etc. Among them, the overcooking stage can be further subdivided into multiple overcooking stages such as a first-level overcooking stage, a second-level overcooking stage, or a mild overcooking stage. In practical applications, it can also be divided according to levels such as a mild overcooking stage, a moderate overcooking stage, and a severe overcooking stage. Different cooking stages may produce different types of gas components, and for the user's intuitive feeling, it is to smell different odors. The amount and concentration of the smoke generated in different cooking stages can be different.

[0041] Furthermore, the preset machine learning model can be preset in the range hood and is called by the control module when implementing this embodiment to predict the amount of smoke and the current cooking stage according to the ambient data collected in step S101.

[0042] Furthermore, the working state of the range hood can include operating states such as shutdown, standby, and startup. Among them, in the startup state, the working state can further include operating states under different exhaust intensities. For example, when it is shutdown and standby, the fan of the range hood is stationary. Also for example, when it is startup, the fan of the range hood runs, and as the exhaust intensity increases, the rotation speed of the fan increases accordingly.

[0043] In a specific implementation, the preset machine learning model can include a first model for predicting the amount of smoke according to the ambient data. Specifically, the first model can receive the gas data, the ambient temperature, and the ambient humidity to comprehensively consider and predict the amount of smoke in the cooking area.

[0044] Furthermore, the preset machine learning model can include a second model for predicting the current cooking stage according to the ambient data. Specifically, the second model can receive the gas data, the ambient temperature, and the ambient humidity to comprehensively consider and predict the current cooking stage of the dishes in the cooking area.

[0045] The first model and the second model can be two independent models, and the two models operate independently to obtain their respective prediction results. For either the first model or the second model, the model can be constructed based on a machine learning algorithm, which can be, for example, a decision tree, naive Bayes classification, least squares regression, logistic regression, support vector machine, neural network, deep learning, multilayer perceptron (MLP), random forest algorithm, Extreme Gradient Boosting Decision Tree (XGBoost), and K-Nearest Neighbor (KNN) classification algorithm, etc.

[0046] Further, step S102 may include the steps of: inputting the environmental data into the first model to obtain a first prediction result, where the first prediction result includes the prediction result of the smoke volume; inputting the environmental data into the second model to obtain a second prediction result, where the second prediction result includes the prediction result of the current cooking stage.

[0047] The action of inputting environmental data into the first model to obtain a first prediction result and the action of inputting environmental data into the second model to obtain a second prediction result can be executed synchronously or asynchronously. When executed asynchronously, the order of execution of the two actions can be interchanged.

[0048] Thus, the preset machine learning model combines a smoke volume prediction algorithm and an overcooking detection algorithm, and can accurately control the working state of the range hood and achieve timely monitoring of overcooking according to the prediction results. Specifically, the first model is dedicated to predicting the size of the smoke volume to achieve real-time detection of the smoke volume, so as to accurately control the working state of the range hood. Further, the second model is dedicated to predicting the current cooking stage to achieve timely detection of the overcooking stage.

[0049] In a specific implementation, the first model and the second model can be constructed using different machine learning algorithms. For example, the first model can be constructed using a long short-term memory neural network, and the second model can be constructed using a support vector machine.

[0050] In some embodiments, in the model construction and training stage, multiple machine learning algorithms can be used to construct models respectively, each model is trained with the same training set, and the model with the highest prediction accuracy for the smoke volume is determined as the first model.

[0051] Similarly, multiple machine learning algorithms can be used to build models respectively, and these models are also trained and verified with the training set. The model with the highest prediction accuracy for the current cooking stage is determined as the second model.

[0052] Furthermore, when selecting appropriate models as the first model and the second model respectively, there is no correlation in the selection logic. In other words, the construction and selection of the first model and the second model can be achieved with the prediction accuracy as the primary measurement criterion.

[0053] Thus, the first model and the second model can be independently constructed and trained so that the prediction results of each model conform to their respective model construction objectives. Furthermore, constructing the first model and the second model respectively with the prediction accuracy as the criterion is conducive to improving the accuracy of the first prediction result and the second prediction result.

[0054] In a specific implementation, step S103 may include the steps: if the first prediction result indicates that the amount of smoke generated by cooking is zero, adjust the working state of the range hood to the shutdown or standby state; if the first prediction result indicates that the amount of smoke generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the normal cooking stage, adjust the rotation speed of the fan in the range hood according to the first prediction result; if the first prediction result indicates that the amount of smoke generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the overcooking stage, control the fan of the range hood to operate at the maximum rotation speed.

[0055] Specifically, in the shutdown state, the range hood can be completely powered off to better save energy and protect the environment. Further, in the standby state, the range hood can be in the sleep mode, and at this time, the control module can operate with low power consumption to reduce the overall energy consumption of the range hood.

[0056] Furthermore, in the shutdown and standby states, the control module can be periodically awakened to execute this implementation scheme to ensure timely response to the real-time change of the smoke amount in the cooking area and ensure timely detection of the overcooking phenomenon.

[0057] Further, in the normal cooking stage, the rotation speed of the fan can be adjusted to operate at the corresponding speed according to the specific value of the predicted smoke amount in the first prediction result. For example, a large gas volume (i.e., when the value of the first prediction result is large) corresponds to a high rotation speed, a medium gas volume corresponds to a medium rotation speed, and a small gas volume corresponds to a low rotation speed. Thus, it is ensured that the exhaust force of the range hood can extract most of the smoke amount with as low power consumption as possible at this time, providing a better cooking experience for the user.

[0058] Further, in the overcooking stage, indicating a large amount of smoke or at least a high concentration of oil fumes at this time, it is preferably to control the fan to run at full speed to exhaust the oil fumes as soon as possible with the maximum exhaust force to ensure cooking safety.

[0059] Further, the actions of step S101 and step S102 can be continuously executed. For example, after detecting the overcooking phenomenon and controlling the fan to run at the maximum speed, the environmental data can be continuously monitored and step S102 can be executed until the second prediction result indicates that the current cooking stage has returned to the normal cooking stage and / or the first prediction result indicates that the amount of smoke in the current cooking area is zero, then the rotation speed of the fan is restored to be adjusted according to the first prediction result.

[0060] Thus, by combining the first prediction result and the second prediction result, the real-time cooking environment is comprehensively predicted to determine the most suitable working state of the range hood at present, and then the rotation speed of the fan is accurately controlled so that the range hood works in the most suitable working state for the current cooking environment.

[0061] In a specific implementation, referring to Figure 2 , the training process of the preset machine learning model can include the following steps:

[0062] Step S201, obtaining a training set, where the training set includes environmental data and corresponding calibration results (or called labels) obtained when different ingredients are used with various cooking methods and at multiple cooking stages. Among them, the multiple cooking stages at least include the normal cooking stage and the overcooking stage, and the calibration results include the standard amount of smoke and the standard cooking stage;

[0063] Step S202, training the preset machine learning model based on the training set until the prediction accuracy rate of the preset machine learning model reaches a preset threshold.

[0064] Specifically, the training set can include multiple groups of ingredients, some of which can include a single type of ingredient, and some of which can include a mixed type of ingredient, so as to enrich the diversity of the training set.

[0065] Further, the cooking method can correspond to the cooking action.

[0066] Further, the specific value of the preset threshold can be set according to user needs. For example, it can take values from 87% to 99%. The preset threshold can be used to measure the user's tolerance for the prediction error rate of the model. When the first model and the second model are respectively selected from multiple candidate models, the preset thresholds of each candidate model can be the same or different.

[0067] Thus, by enriching the diversity of training samples in the training set, the preset machine learning model can also have excellent performance in complex environments and can accurately identify the smoke volume and the current cooking stage for different ingredients and different cooking stages.

[0068] In some embodiments, the prediction accuracy of the preset machine learning model can be obtained by testing with a test set after initially training the model based on the training set. Among them, the process of obtaining the test set can refer to the generation process of the training set, and the difference is that the test set is not input to the preset machine learning model during the model construction stage.

[0069] In some embodiments, environmental data of a large number of gases with different volumes can be collected to construct a preset machine learning model (for example, the first model) for detecting the gas volume, and the gas volume can characterize the smoke volume. Further, when constructing the training set, different volumes of gases generated by different cooking methods and different cooking processes can be consciously collected to enrich the training samples in the training set. When collecting environmental data, real-time calibration can be arranged manually or based on a gas analyzer to determine the calibration result (for example, the standard smoke volume) corresponding to the environmental data.

[0070] In some embodiments, environmental data during overcooking and normal cooking can be collected to construct a preset machine learning model (for example, the second model) for detecting the current cooking stage. As many different types of foods as possible can be consciously overcooked, and some foods that are often cooked together can be mixed, and the gases generated during the cooking process can be collected. Then, a gas analyzer is used to analyze the gas components in the collected gas, for example, to analyze the main gas components that will appear in the gas during overcooking. According to each gas component generated during overcooking (i.e., the gas component to be detected), the dynamic heating temperature of the gas detector (for example, the VOC sensor) is adjusted so that the VOC sensor is more sensitive to the gas component to be detected. The adjusted VOC sensor is installed at a suitable position on the cooker to be prepared to receive data during overcooking.

[0071] Next, design a list of ingredients for conducting overcooking experiments, conduct overcooking experiments under the cooker, and collect gas data during the cooking process based on the VOC sensor. At the same time, arrange calibrators to conduct real-time calibration on site. For example, when there is no smell due to overcooking, the calibrator can calibrate that the current is the normal cooking stage, and when the smell due to overcooking appears for the first time, overcooking is calibrated.

[0072] Further, the overcooking stage can be further refined into a slight burnt smell stage, a moderate burnt smell stage, and a severe burnt smell stage. The real-time calibration results of the calibrator, the corresponding gas data, and the environmental temperature and humidity can generate the training set.

[0073] In some embodiments, multiple experiments can be conducted repeatedly. According to the collected environmental data, the data of the last experiment is used as the test set, and the remaining data is divided into a training set and a validation set for generating a preset machine learning model. For example, a preset machine learning model using a specific algorithm is constructed based on the training set, and then the model parameters of the preset machine learning model are verified and adjusted using the validation set. Finally, the prediction accuracy of the preset machine learning model is tested using the test set.

[0074] In a specific implementation, the training process of the first model may include: obtaining a first set, where the first set includes environmental data and corresponding standard smoke amounts respectively obtained when different ingredients are subjected to various cooking methods and at multiple cooking stages; training the first model based on the first set until the prediction accuracy of the first model reaches a first threshold.

[0075] Specifically, the first set can be constructed based on the training set. For example, the first set can be formed based on the environmental data and the corresponding standard smoke amounts in the training set.

[0076] Furthermore, the first model can be selected from multiple candidate models, where different candidate models are constructed using different algorithms.

[0077] For example, each candidate model can be trained based on the first set until the prediction accuracy of each candidate model reaches the first threshold. Then, the candidate model with the highest prediction accuracy among the multiple trained candidate models is selected as the first model. Thus, using the candidate model with the best prediction performance as the final first model in actual use is beneficial to improving the prediction accuracy of the smoke amount in the actual application stage.

[0078] Furthermore, the specific value of the first threshold can be preset by the manufacturer of the range hood or can be flexibly adjusted according to user needs. For example, the first threshold can take values from 85% to 99%.

[0079] Similarly, the training process of the second model may include: obtaining a second set, where the second set includes environmental data and corresponding standard cooking stages respectively obtained when different ingredients are subjected to various cooking methods and at multiple cooking stages; training the second model based on the second set until the prediction accuracy of the second model reaches a second threshold.

[0080] Specifically, the second set can be constructed based on the training set. For example, the second set can be formed based on the environmental data and the corresponding standard cooking stages in the training set.

[0081] Furthermore, the second model can be selected from multiple candidate models, where different candidate models are constructed using different algorithms.

[0082] For example, each candidate model can be trained based on the second set respectively, and the candidate model among the candidate models whose prediction accuracy rate first reaches the second threshold is determined as the second model.

[0083] Furthermore, the specific value of the second threshold can be preset by the manufacturer of the range hood or can be flexibly adjusted according to user needs. For example, the second threshold can take values from 85% to 99%.

[0084] Thus, different sets are constructed corresponding to the specific dimensions of the prediction results of the two models for targeted training, so that the prediction accuracy rate of the first prediction result of the first model in the smoke volume dimension meets the expectations, and the prediction accuracy rate of the second prediction result of the second model in the cooking stage dimension meets the expectations. Furthermore, by enriching the diversity of the training samples in the two sets, the first model and the second model can also have better performance in complex environments and can accurately identify the smoke volume and the current cooking stage for different ingredients and different cooking stages.

[0085] In some embodiments, the numerical values of the first threshold and the second threshold can be different. Thus, the expectations for the respective prediction accuracies of the two models can be flexibly set according to needs, so that the trained first model and second model meet user expectations. For example, the first threshold can be greater than the second threshold, and vice versa.

[0086] In a specific implementation, the ambient temperature in the cooking area can be determined based on multiple candidate temperatures, and the multiple candidate temperatures are respectively collected from different positions in the cooking area. For example, both the range hood and the stove can have sensors for temperature measurement to respectively collect the candidate temperatures at different positions.

[0087] In some embodiments, the ambient temperature can be the average value of multiple candidate temperatures.

[0088] In some embodiments, the ambient temperature can be obtained by weighted averaging based on multiple candidate temperatures. Among them, the weights of different positions are different. For example, the candidate temperature collected closer to the burner head of the cooking appliance has a greater corresponding weight.

[0089] Thus, calculating the ambient temperature comprehensively based on the candidate temperatures at multiple positions is beneficial to improving the detection accuracy of the ambient temperature.

[0090] In a specific implementation, the ambient humidity in the cooking area can be determined based on multiple candidate humidities, and the multiple candidate humidities are respectively collected from different positions in the cooking area. For example, both the range hood and the stove have sensors for humidity measurement to respectively collect the candidate humidities at different positions.

[0091] In some embodiments, the ambient humidity can be the average value of multiple candidate humidities.

[0092] In some embodiments, the ambient humidity can be obtained by weighted averaging of multiple candidate humidities. Among them, the weights of different positions are different. For example, the candidate humidity collected closer to the air-permeable part of the cooking utensil has a greater corresponding weight.

[0093] Thus, the ambient humidity is comprehensively calculated based on the candidate humidities at multiple positions, which is beneficial to improving the detection accuracy of the ambient humidity.

[0094] As described above, by adopting this implementation scheme, the working state of the range hood can be automatically and intelligently adjusted precisely according to the current cooking environment, and the overall performance of the range hood can be actively made to meet the user's real-time cooking needs without manual adjustment by the user, improving the safety of the cooking process. Specifically, the smoke volume and the current cooking stage are comprehensively determined according to the gas data and the ambient temperature and humidity data to more precisely adjust the working state of the range hood. Among them, the working state can include the operating parameters of the fan (for example, the rotational speed of the fan), and the current cooking stage can include the normal cooking stage and the overcooking stage. Thus, the rotational speed of the fan can be more precisely controlled, which is beneficial to preventing food waste and burning of cooking utensils, making the cooking process safer.

[0095] Figure 3 It is a schematic diagram of a typical application scenario of a range hood 3 according to an embodiment of the present invention.

[0096] Specifically, referring to Figure 3 , the range hood 3 described in this embodiment may include: a main body 31; a control module 32 disposed in the main body 31, and the control module 32 is used to execute the above Figure 1 and Figure 2 shown range hood adaptive control method; a collection module 33 disposed in the main body 31 and communicating with the control module 32, and the collection module 33 is used to collect the environmental data and transmit it to the control module 32.

[0097] More specifically, a flue can be provided in the main body 31, and the flue has an air inlet towards the cooking area and an opposite air outlet, and the cooking fumes generated in the cooking area enter the flue from the air inlet and then are discharged from the air outlet. Further, a cooking appliance 4 can be provided in the cooking area.

[0098] In some embodiments, the control module 32 can be the main control board of the range hood 3, and the main control board is used to adjust the exhaust strength, lighting, power on and off, etc. of the range hood 3. Further, the cooking appliance 4 can also have a main control board for adjusting the heating power, firepower size, etc. of the cooking appliance 4.

[0099] In some embodiments, the acquisition module 33 may be disposed in the flue to acquire gas data of the gas entering the flue. Further, the acquisition module 33 may also acquire the ambient temperature and ambient humidity of the cooking area.

[0100] As described above, the adaptive control method for the range hood provided by this implementation scheme can accurately judge the real-time smoke volume and the current cooking stage, so that the range hood 3 can intelligently and accurately adjust its own working state, so that the working state of the range hood 3 always conforms to the real-time cooking environment. Further, the range hood 3 can operate while maintaining a working state consistent with the real-time cooking environment, which is beneficial to preventing food waste and burning cooking utensils, making the cooking process safer.

[0101] In a specific implementation, with continued reference to Figure 3 , the range hood 3 may further include a fan 34 disposed on the main body 31, and the control module 32 may adjust the rotation speed of the fan 34 according to the prediction result.

[0102] Specifically, the fan 34 may be disposed in the flue to promote gas flow in the flue.

[0103] Further, the greater the rotation speed of the fan 34, the greater the exhaust force of the range hood 3.

[0104] Thus, by combining the automatic fan speed control algorithm and the overcooking detection algorithm, the rotation speed of the fan 34 can be accurately controlled and overcooking detection can be achieved. Specifically, the automatic fan speed control algorithm may include automatically controlling the rotation speed of the fan 34 based on the predicted smoke volume result of a preset machine learning model, and the overcooking detection algorithm may include detecting whether there is an overcooking phenomenon based on the predicted result of the current cooking stage of a preset machine learning model.

[0105] In a specific implementation, with continued reference to Figure 3 , the acquisition module 33 may include a gas detector 331. Thus, the gas data in the cooking area is acquired based on the gas detector 331.

[0106] Specifically, the gas detector 331 may include a volatile organic compounds (VOC) sensor. The VOC sensor reacts strongly to specific types of gases at different temperatures. Utilizing this characteristic enables the gas detector 331 used in this implementation scheme to detect multiple gas components within a certain temperature range. For example, at a specific working temperature, the VOC sensor reacts with specific types of gas components in the air and outputs a resistance value, which can be used to characterize the concentration of the specific type of gas component.

[0107] During the operation of the gas detector 331, the operating temperature of the gas detector 331 is adjusted, and candidate gas data respectively collected by the gas detector 331 at multiple operating temperatures are received. Among them, the candidate gas data associated with different operating temperatures are used to characterize the concentrations of different gas components to be detected generated by cooking. The gas data is generated based on the received multiple candidate gas data.

[0108] During the operation of the gas detector 331, the operating temperature of the gas detector 331 is adjusted according to a candidate value selected from a preset set of operating temperatures, where the preset set of operating temperatures includes multiple candidate values of the operating temperature. By reasonably designing the preset set of operating temperatures, the gas detector 331 can accurately detect various gas components to be detected.

[0109] The candidate values in the preset set of operating temperatures can be determined according to the sensitivity of the gas detector 331 to the gas components to be detected when it operates at the candidate values. For example, various gas components mainly generated by various food ingredients during overcooking can be determined through preliminary experiments. Taking this gas component as the target, the operating temperature of the VOC sensor is adjusted to test at which operating temperature the VOC sensor has the highest sensitivity (i.e., the sensitivity) to this type of gas component, and the time required to detect this gas component is measured. The operating temperature with the highest sensitivity is added to the preset set of operating temperatures as a candidate value that can be selected when implementing this embodiment. Thus, the gas detector 331 can detect a specific type of gas component at the operating temperature that is most sensitive to this type of gas component, thereby ensuring the accuracy of the gas data, which is beneficial to ensuring the accuracy of the final prediction result.

[0110] Furthermore, the gas detector 331 communicates with the control module 32 to send the gas data to the control module 32. For example, the gas detector 331 can be arranged in the flue of the range hood 3 so as to detect the gas data by the way when the range hood 3 exhausts.

[0111] Furthermore, the surface of the gas detector 331 can be covered with a protective film to reduce the influence of moisture generated during cooking on the detection accuracy.

[0112] Furthermore, the gas detector 331 and the control module 32 can communicate in a wired or wireless manner.

[0113] In a specific implementation, the acquisition module 33 can include a temperature and humidity sensor 332. Thus, based on the temperature and humidity sensor 332, the temperature data and humidity data in the cooking area are acquired.

[0114] Specifically, the temperature and humidity sensor 332 can be set at any position within the cooking area. For example, it can be set in the flue of the range hood 3. Another example is that it can also be externally installed on the range hood 3, such as being set on the cooking stove 4.

[0115] Furthermore, the temperature and humidity sensor 332 communicates with the control module 32 to send the temperature data and humidity data to the control module 32.

[0116] In some embodiments, the temperature and humidity sensor 332 can be a single sensor that can simultaneously collect temperature data and humidity data. For example, the temperature and humidity sensor 332 can be an environmental sensor, set in the flue or near the burner of the cooking stove. The environmental sensor sends the collected temperature data and humidity data to the control module 32 together.

[0117] In some embodiments, the temperature and humidity sensor 332 can be a collection of multiple sensors. A part of the multiple sensors is used to collect temperature data, and the remaining part is used to collect humidity data. For example, the temperature and humidity sensor 332 can include an infrared sensor, an NTC (Negative Temperature Coefficient) temperature sensor, a humidity sensor, etc. The infrared sensor can be set in the flue to collect the ambient temperature, the NTC sensor can be set on the cooking stove 4 to collect the ambient temperature, and the humidity sensor can be set in the flue to collect the ambient humidity.

[0118] In response to receiving the ambient temperatures transmitted by the infrared sensor and the NTC sensor respectively, the control module 32 can determine the average value of these two ambient temperatures as the ambient temperature input into the preset machine learning model.

[0119] In a specific implementation, the control module 32 also communicates with at least one auxiliary sensor. The auxiliary sensor is externally installed on the range hood 3 and is separately set at different positions in the cooking area from the range hood 3. The auxiliary sensor is used to collect at least a part of the ambient data and transmit it to the control module 32.

[0120] Specifically, the collection module 33 externally installed on the range hood 3 can be collectively referred to as the auxiliary sensor. For example, referring to Figure 3 , the temperature and humidity sensor 332 set on the cooking stove 4 can be regarded as an auxiliary sensor.

[0121] Thus, the collection result of the auxiliary sensor can strengthen and optimize the collection result of the collection module 33, so that the ambient data input into the preset machine learning model can more accurately represent the real-time cooking environment.

[0122] In some embodiments, the acquisition module 33 provided in the range hood 3 can acquire environmental data in all dimensions. For example, a gas detector 331 and an environmental sensor can be simultaneously provided in the flue. In this example, the sensing results of the auxiliary sensors can characterize the real-time cooking environment from different positions, thereby optimizing the acquisition results of the acquisition module 33. For example, the control module 32 can average the environmental temperature collected by the temperature and humidity sensor provided on the cooking stove 4 and the environmental temperature collected by the environmental sensor provided in the flue as the final environmental temperature input to the preset machine learning model.

[0123] In some embodiments, the acquisition module 33 provided in the range hood 3 can acquire environmental data in some dimensions, and the environmental data in other dimensions is obtained based on auxiliary sensors. Thus, the auxiliary sensors can reinforce the acquisition results of the acquisition module 33. Combining the environmental data obtained by the acquisition module 33 and the auxiliary sensors is beneficial to enrich the dimensional diversity of the input data of the preset machine learning model. For example, a temperature and humidity sensor 332 can be provided on the cooking stove 4 to collect the environmental temperature / environmental humidity. These data and the gas data collected by the acquisition module 33 (such as the gas detector 331) provided in the range hood 3 are input into the preset machine learning model to obtain a prediction result.

[0124] In a specific implementation, with continued reference to Figure 3 , the range hood 3 may further include a communication module 35, and the control module 32 and the gas detector 331 can communicate with each other through the communication module 35. For example, the communication module 35 can be provided on the main body 31 of the range hood 3 to facilitate signal transmission between the control module 32 and the gas detector 331.

[0125] Furthermore, the control module 32 and the temperature and humidity sensor 332 can also communicate with each other through the communication module 35. In the embodiment where the temperature and humidity sensor 332 is provided on the cooking stove 4, the cooking stove 4 can also be integrated with a communication module (not shown in the figure) and communicate with the control module 32 through the communication module 35 to send the environmental temperature and environmental humidity collected by the temperature and humidity sensor 332 to the control module.

[0126] In some embodiments, the communication module 35 can also communicate with the user's device terminal to timely inform the user of the prediction result of the preset machine learning model. For example, the device terminal can include a mobile phone, an IPAD, a home monitoring device, etc.

[0127] It should be noted that Figure 3Only an exemplary display of the possible installation positions of the control module 32, the acquisition module 33, the fan 34, and the communication module 35 on the range hood 3 and the cooking appliance 4 is shown. In actual applications, the relative positions of the modules and their specific installation positions on the range hood 3 and the cooking appliance 4 can be adjusted as needed. The modules can be independent of each other or integrated on the same chip or into the same functional module. For example, the control module 32 and the communication module 35 can be integrated together.

[0128] In a specific implementation, with continued reference to Figure 3 , the range hood 3 may further include a storage module 36, which is disposed on the main body 31 and communicates with the control module 32. The storage module 36 is used to store a preset machine learning model.

[0129] Specifically, the preset machine learning model can be pre-deployed in the storage module 36 so that the control module 32 can call it when needed.

[0130] For example, the storage module 36 can store a first model and a second model. When the control module 32 executes the steps S102 described in the above Figure 1 shown embodiment, it accesses the storage module 36 to call the first model and the second model. Then, the environmental data obtained in step S101 is input into the first model and the second model respectively to obtain a first prediction result and a second prediction result.

[0131] Thus, the preset machine learning model is pre-installed in the range hood 3 and is called when executing this implementation scheme to realize the prediction of the smoke volume and the current cooking stage.

[0132] It should be noted that Figure 3 only an exemplary display of the possible installation position of the storage module 36 on the range hood 3 is shown. In actual applications, its specific installation position on the range hood 3 and the cooking appliance 4 and the relative position relationship between the storage module 36 and other modules on the range hood 3 can be adjusted as needed. The control module 32 and the storage module 36 can be integrated together or be two independent components.

[0133] In a typical application scenario, with reference to Figure 3 , a trigger button (not shown in the figure) can be provided on the display interface 37 of the range hood 3. When the user clicks the button, the adaptive control function of the range hood is triggered. Alternatively, the user can remotely trigger this function through a smart terminal that communicates with the range hood 3. In response to the function being triggered, the control module 32 executes the above Figure 1 shown range hood adaptive control method.

[0134] Specifically, with reference to Figure 4 , the control module 32 can control the acquisition module 33 to work to obtain environmental data.

[0135] For example, the control module 32 controls the gas detector 331 to operate to collect gas data during the cooking process. Further, the gas detector 331 sends the collected gas data to the control module 32.

[0136] For another example, the control module 32 controls the temperature and humidity sensor 332 disposed on the cooking appliance 4 to operate through the communication module 35 to collect the ambient temperature and ambient humidity of the cooking area. Further, the temperature and humidity sensor 332 sends the collected ambient temperature and humidity to the control module 32 through the communication module of the cooking appliance 4 and the communication module 35.

[0137] Further, in response to receiving the environmental data, the control module 32 inputs the environmental data into the first model and the second model respectively. Among them, the first model predicts the amount of smoke generated in the cooking area currently due to the cooking behavior according to the environmental data, and the second model predicts the current cooking stage of the cooking area according to the environmental data.

[0138] Further, according to the first prediction result of the first model, the control module 32 determines whether there is a cooking behavior in the cooking area currently. For example, if the first prediction result indicates that the amount of smoke generated by cooking is zero, indicating that there is no cooking behavior, the control module 32 controls the fan 34 to be stationary or stop running. For another example, if the first prediction result indicates that the amount of smoke generated by cooking is non-zero, indicating that there is a cooking behavior.

[0139] According to the second prediction result of the second model, the control module 32 determines the current cooking stage. Further, the monitoring of the current cooking stage can be carried out continuously.

[0140] Further, when the first prediction result indicates that there is a cooking behavior, the control module 32 determines whether there is an overcooking phenomenon according to the second prediction result. For example, if the second prediction result indicates that the current is in the normal cooking stage, the control module 32 adjusts the rotation speed of the fan 34 according to the amount of smoke predicted by the first model. For another example, if the second prediction result indicates that the current is in the overcooking stage, the control module 32 controls the fan 34 to run at the maximum rotation speed until the second prediction result of the second model returns to the normal cooking stage or the first prediction result of the first model indicates that the cooking behavior ends.

[0141] Although the specific embodiments have been described above, these embodiments are not intended to limit the scope of the present disclosure, even in the case where a single embodiment is described only with respect to a specific feature. The feature examples provided in the present disclosure are intended to be illustrative rather than restrictive, unless otherwise stated. In a specific implementation, the technical features of one or more dependent claims can be combined with the technical features of the independent claim, and the technical features from the corresponding independent claims can be combined in any appropriate manner rather than only through the specific combinations listed in the claims.

[0142] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. An adaptive control method for a range hood, characterized in that, Including: Obtain environmental data, where the environmental data includes gas data for characterizing the concentration of at least one gas component generated by cooking, the environmental temperature and environmental humidity of the cooking area; Input the environmental data into a preset machine learning model and obtain a prediction result, where the preset machine learning model is used to predict the current cooking stage and the amount of smoke generated by cooking based on the environmental data; Adjust the working state of the range hood according to the prediction result.

2. The method according to claim 1, wherein The preset machine learning model includes: a first model for predicting the amount of smoke based on the environmental data; a second model for predicting the current cooking stage based on the environmental data; The step of inputting the environmental data into the preset machine learning model and obtaining a prediction result includes: Input the environmental data into the first model to obtain a first prediction result, where the first prediction result includes the prediction result of the amount of smoke; Input the environmental data into the second model to obtain a second prediction result, where the second prediction result includes the prediction result of the current cooking stage.

3. The method according to claim 2, characterized in that, The step of adjusting the working state of the range hood according to the prediction result includes: If the first prediction result indicates that the amount of smoke generated by cooking is zero, adjust the working state of the range hood to the shutdown or standby state; If the first prediction result indicates that the amount of smoke generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the normal cooking stage, adjust the rotation speed of the fan in the range hood according to the first prediction result; If the first prediction result indicates that the amount of smoke generated by cooking is non-zero, and the second prediction result indicates that the current cooking stage is the overcooking stage, control the fan of the range hood to operate at the maximum rotation speed.

4. The method according to claim 2, wherein The first model and the second model are constructed using different machine learning algorithms.

5. The method according to claim 1, characterized in that The training process of the preset machine learning model includes: Obtain a training set, where the training set includes environmental data and corresponding calibration results obtained when different ingredients are cooked by various cooking methods and at multiple cooking stages. The multiple cooking stages at least include the normal cooking stage and the overcooking stage, and the calibration results include the standard amount of smoke and the standard cooking stage; Train the preset machine learning model based on the training set until the prediction accuracy of the preset machine learning model reaches a preset threshold.

6. The method according to claim 5, wherein The preset machine learning model includes: a first model for predicting the amount of smoke based on the environmental data; a second model for predicting the current cooking stage based on the environmental data; The training process of the first model includes: Obtain a first set, where the first set includes environmental data and corresponding standard amounts of smoke obtained when different ingredients are cooked by various cooking methods and at multiple cooking stages; Train the first model based on the first set until the prediction accuracy of the first model reaches a first threshold; The training process of the second model includes: Obtain a second set, where the second set includes environmental data and corresponding standard cooking stages respectively obtained when different cooking ingredients are subjected to various cooking methods and at multiple cooking stages. Train the second model based on the second set until the prediction accuracy of the second model reaches a second threshold.

7. The method according to claim 6, characterized in that The numerical values of the first threshold and the second threshold are different.

8. The method according to claim 1, wherein The ambient temperature of the cooking area is determined according to a plurality of candidate temperatures, and the plurality of candidate temperatures are respectively collected from different positions of the cooking area; and / or, the ambient humidity of the cooking area is determined according to a plurality of candidate humidities, and the plurality of candidate humidities are respectively collected from different positions of the cooking area.

9. An oil fume machine, characterized in that, Comprising: A body (31); A control module (32) disposed on the body (31), and the control module (32) is used to execute the method according to any one of claims 1 to 8 above. An acquisition module (33) disposed on the body (31) and communicating with the control module (32), and the acquisition module (33) is used to acquire the environmental data and transmit it to the control module (32).

10. The range hood according to claim 9, characterized in that, Further comprising: A fan (34) disposed on the body (31), and the control module (32) adjusts the rotation speed of the fan (34) according to the prediction result.

11. The range hood according to claim 9, wherein, Further comprising: A storage module (36) disposed on the body (31) and communicating with the control module (32), and the storage module (36) is used to store the preset machine learning model.

12. The range hood according to claim 9, wherein, The control module (32) also communicates with at least one auxiliary sensor, the auxiliary sensor is externally disposed on the range hood and the range hood and the auxiliary sensor are respectively disposed at different positions in the cooking area, and the auxiliary sensor is used to collect at least a part of the environmental data and transmit it to the control module (32).

13. The range hood according to claim 9, characterized in that, The acquisition module (33) includes: a gas detector (331) and a temperature and humidity sensor (332).

Images